ChatGPT

I have been hearing a lot about ChatGPT recently. ChatGPT was launched in November 2022 by OpenAI, and in its own words it "is a language model developed by OpenAI. It is a machine learning model that can generate text, such as writing essays, article, stories and also it can generate text based on given context. ChatGPT is based on the GPT (Generative Pre-training Transformer) architecture, which was trained on a massive amount of data, allowing it to generate text that is often indistinguishable from text written by a human. This model is one of the most advanced language model and used for various NLP tasks."

My most recent experience with OpenAI was with DALL-E. DALL-E is based on the GPT-3 architecture (as an aside, the GPT-3 architecture is also used by Duolingo for French grammar corrections) and can generate images based on text inputs and descriptions. It definitely comes up with some very interesting results. In my experience, the more detailed the input/request, the better the resulting images was. In the short term future it can be very useful as a prototyping/storyboarding tool, and in the longer term may begin to compete with humans for some jobs. Perhaps the ability to wrangle and control DALL-E could become a very useful skillset (or microcredential – ha!).

I finally bit the bullet to try ChatGPT when I received an email from a colleague today. Working in tertiary education, many people have been discussing this recently and wondering how we will ever be able to detect whether something was written by AI or a human being. The email, with names removed:

Hi All,

Here’s an article worth reading… or maybe scary to read 🤯

https://www.theguardian.com/technology/2023/jan/12/college-student-claims-app-can-detect-essays-written-by-chatbot-chatgpt 

If someone had the idea of creating an app to ‘detect’ text written by a chat bot, it’s because such a practice is more widespread than we’d like to think….

I logged into my OpenAI account and quickly found the free research preview. The conversation continued as follows:

What sort of things can you do?

I am a machine learning model that can perform a variety of tasks, such as:

• answering questions (like the one you just asked)

• providing explanations

• Generating text, such as stories, articles, and poetry

• Translating text from one language to another

• Summarizing text

• And much more. Let me know if there's something specific you would like me to do and I will let you know if it is within my capabilities.

Could you generate a response to an email I have received that references a news article?

I'd be happy to help you generate a response to an email. Please provide me with the email and the specific news article that it references, and let me know what kind of response you're looking for. Keep in mind that if the content of the email contain sensitive information or have specific company guideline, it would be best to not share it.

Also, please keep in mind that as a machine learning model, my responses may not be perfect, and it's always a good idea to proofread any text generated by me before sending it.

After then copying the email and the body text of the article, I was met with the following response:

Hi All,

Thanks for sharing this interesting article about the new app, GPTZero, that can detect whether text is written by ChatGPT. The concern of AI being used to cheat in exams is certainly a valid one, and it's commendable that the developer, Edward Tian, is trying to address this issue.

However, it's also important to keep in mind that technology is always evolving, and it's likely that there will be ongoing effort to develop new ways to cheat the system. As Professor Walsh mentioned, it will likely be an "arms race" between those developing anti-plagiarism technology and those looking to evade it.

Overall, it will be interesting to see how this app develops and whether it will have a significant impact on academic integrity.

Overall, this was quite an impressive response (time will tell if my colleagues also think so!). I was particularly impressed by the ability of ChatGPT to gain the context of the quotations in the article, and then paraphrase them appropriately in the response. It seems to me to be an almost by-the-book response to a generic article summary task that I have seen time and time again. Perhaps this would be a good tool for students to use for practice to gain mastery over their learning? This is only one very small use case, I wonder what the effects of passing an entire written task through ChatGPT and asking for feedback would be? To quote ChatGPT, "it will be interesting to see how this [] develops".